The present invention relates to a temperature sensor for measuring a substrate temperature during thermal processing.
In rapid thermal processing (RTP), a substrate is heated quickly and uniformly to a high temperature, such as 4000 Celsius (C) or more, to perform a fabrication step such as annealing, cleaning, chemical vapor deposition, oxidation, or nitration. For example, a thermal processing system, such as the RTP tool available from Applied Materials, Inc., under the trade name "Centura.sup.R ", may be used to perform metal annealing at temperatures of 400.degree. C. to 500.degree. C., titanium silicide formation at temperatures around 650.degree. C., or oxidation or implant annealing at temperatures around 1000.degree. C.
The temperature of the substrate must be precisely controlled during these thermal processing steps to obtain high yields and process reliability, particularly given the submicron dimension of current semiconductor devices. For example, to fabricate a dielectric layer 60-80 angstroms (.ANG.) thick with a uniformity of +/-2 .ANG., a typical requirement in current device structures, the temperature in successive processing runs cannot vary by more than a few .degree. C. from the target temperature. To achieve this level of temperature control, the temperature of the substrate is measured in real time and in situ.
Optical pyrometry is a technology that is used to measure substrate temperatures in RTP systems. Pyrometry exploits a general property of objects, namely, that objects emit radiation with a particular spectral content and intensity that is characteristic of their temperature. Thus, by measuring the emitted radiation, the object's temperature can be determined. A pyrometer measures the emitted radiation intensity and performs the appropriate conversion to obtain the substrate temperature. The relationship between spectral intensity and temperature depends on the spectral emissivity of the substrate and the ideal blackbody intensity-temperature relationship, given by Planck's law: ##EQU1##
where C.sub.1 and C.sub.2 are known constants, .lambda. is the radiation wavelength of interest, and T is the substrate temperature measured in .degree. K. The spectral emissivity .epsilon.(.lambda.,T) of an object is the ratio of its emitted spectral intensity I(.lambda.,T) to that of a black body at the same temperature I.sub.B (.lambda.,T). That is, ##EQU2##
Since C.sub.1 and C.sub.2 are known constants, under ideal conditions, the temperature of the substrate can be accurately determined if .epsilon.(.lambda.,T) is known.
The emissivity of a substrate depends on many factors, including the characteristics of the wafer itself (e.g., temperature, surface roughness, doping level of various impurities, material composition and thickness of surface layers), the characteristics of the process chamber, and the process history of the wafer. Therefore, a priori estimation of substrate emissivity cannot provide a general purpose pyrometric temperature measurement capability. Consequently, the emissivity of the substrate needs to be measured in situ. Furthermore, any uncertainty in the measured emissivity introduces an uncertainty into the temperature measurement.
To reduce this uncertainty, several techniques have been developed for reducing the effect of substrate emissivity on the temperature measurement. One such technique involves placing a reflector plate beneath the back surface of a target substrate to form a reflecting cavity. If the reflector plate were an ideal reflector, it can be shown that because all of the radiation emitted from the substrate would be reflected back to the substrate, the reflecting cavity would act as an ideal black body. That is, the intensity of the radiation within the reflecting cavity would not be a function of the emissivity of the surface of the substrate. Thus, in the ideal case, the reflecting cavity increases the effective emissivity of the substrate to a value equal to one.
However, because the reflector plate is not an ideal reflector, the effective emissivity of the substrate will be less than one, although it will be higher than the substrate's actual emissivity. Therefore, the radiation intensity measured by a temperature sensor will still depend upon the emissivity of the substrate. Consequently, although variations in the actual emissivity of the substrate will have less impact on the measured temperature, there will be uncertainty in the temperature measurement.
Moreover, different portions of the substrate may have different emissivities. Consequently, if the emissivity of the substrate is measured in only one region, there will be uncertainty in the temperature measurements of other regions of the substrate.
In addition, the transparency of the substrate contributes to the uncertainty in the temperature measurement. A portion of the radiation intended to heat the substrate may instead pass through the substrate into the pyrometer. Since the pyrometer includes this transmitted radiation in the measured intensity, the transmitted radiation results in an artificially high temperature measurement. Assuming that the substrate is a silicon wafer and the pyrometer is sensitive to infrared radiation, this problem will be more acute at lower processing temperatures (e.g., less than 600.degree. C.), where the transmitivity of silicon to infrared radiation is higher.
Another source of uncertainty is electrical noise. If only a small amount of light enters the pyrometer, the signal-to-noise ratio will be low, and the electrical noise will create uncertainty in the temperature measurement.
Another problem is that there are many thermal processing steps which are not compatible with a highly reflective reflector plate. For example, certain thermal processing steps may be corrosive or destructive to such a reflector plate.
Furthermore, even standard thermal processing may cause the reflector plate to become dirty or corroded over time, and thus less reflective. If the reflector plate's reflectivity decreases, the substrate's effective emissivity also decreases. This change in the substrate's effective emissivity changes the intensity of the radiation sampled by the pyrometer, and can create an error in the measured temperature.